Small Object Tracking in LiDAR Point Clouds: Learning the Target-Awareness Prototype and Fine-Grained Search Region

Area of Science:

• Computer Vision
• Robotics
• Artificial Intelligence

Background:

• LiDAR point clouds are crucial for AI systems like autonomous driving and robotics.
• Tracking small objects in LiDAR data is challenging due to sparse points and sensitivity to disturbances.
• Existing methods often overlook the specific difficulties of small object tracking.

Purpose of the Study:

• To develop a robust deep neural network framework for accurate small object tracking in LiDAR point clouds.
• To address the challenges of sparse foreground points and feature map disturbances in small object detection.
• To enhance the performance and robustness of LiDAR-based object tracking systems.

Main Methods:

• A Siamese network is employed for feature extraction.
• The Target-Awareness Prototype Mining (TAPM) module uses masked auto-encoder reconstruction to enhance foreground point salience.
• The Regional Grid Subdivision (RGS) module leverages Vision Transformer and pixel shuffle for improved feature detail and disturbance tolerance.

Main Results:

• The proposed method achieved a mean Success rate of 64.9% under original settings and 60.4% under scaled settings.
• Performance surpassed existing benchmarks by +3.6% and +5.4% respectively.
• Scaling experiments demonstrated the tracker's robustness in handling small objects.

Conclusions:

• The novel deep neural network framework effectively improves small object tracking in LiDAR point clouds.
• The TAPM and RGS modules contribute to enhanced target salience and disturbance tolerance.
• The method shows significant potential for advancing perception capabilities in autonomous systems.